This week in science: A close-up on Jupiter, a new method to test self-driving cars, and AI

This week in science is a review of the most interesting scientific news of the past week.

An image of Jupiter taken by the Juno spacecraft. Credit: J.E.P. Connerney et al., Science (2017).

First results revealed from Juno’s mission to Jupiter

NASA's Juno spacecraft has been in orbit around Jupiter since July 2016. It carries eight scientific instruments designed to study the planet's interior structure, atmosphere, and magnetosphere, in a mission to better understand gas giants and Jupiter itself.

Juno’s first scientific results were published last week in a pair of papers in a special edition of the Science journal. The data was collected by the Jovian Auroral Distributions Experiment (JADE), which detects the electrons and ions associated with Jupiter's auroras, and by the Ultraviolet Imaging Spectrograph (UVS), which examines the auroras in UV light to study Jupiter's upper atmosphere.

Credit: J.E.P. Connerney et al., Science (2017).

Among the surprising features detected by the experiments was a north-south asymmetry in Jupiter's signature bands, which also disappears near its poles, as can be seen in the image above. Furthermore, the measurements of the planet’s gravity field differ significantly from what was expected. As stated by Dr. Scott Bolton, from the Southwest Research Institute:

"[The difference in the gravity field measurements] has implications for the distribution of heavy elements in the interior, including the existence and mass of Jupiter's core."

Both papers can be accessed on the Science website: S.J. Bolton et al., "Jupiter's interior and deep atmosphere: The initial pole-to-pole passes with the Juno spacecraft," Science (2017); J.E.P. Connerney at Space Research Corporation in Annapolis, MD et al., "Jupiter's magnetosphere and aurorae observed by the Juno spacecraft during its first polar orbits," Science (2017).

Via: Phys.org

New method to test self-driving cars cuts 99.9 percent of validation time and costs

Researchers from the University of Michigan have used data from more than 25.2 million miles of real-world driving that involved nearly 3,000 vehicles and volunteers over the course of two years to develop a new method to test autonomous vehicles. If applied in substitution to the current methods of testing, it could save 99.9 percent of testing time and costs.

According to the researchers, for consumers to accept driverless vehicles, it is expected that consumers achieve an 80 percent confidence that those vehicles are 90 percent safer than human drivers. To achieve that, test vehicles using current validation methods would need up to 11 billion miles of real-world or simulated testing, mainly because the most important scenarios they need to tackle are rare. For example, a crash that results in a fatality, on average, occurs just once in every 100 million miles of driving.

But the new accelerated evaluation method proposed breaks down difficult real-world driving situations into components that can be tested or simulated repeatedly. By doing so, self-driving cars could be exposed directly to the most important scenarios to accelerate the validation process. The researchers have estimated that just 1,000 miles of this kind of testing can yield the equivalent of 300,000 to 100 million miles of real-world driving.

For now, the researchers have evaluated only two scenarios: an automated car following a human driver and a human driver merging in front of an automated car. Those are considered the most commonly expected to yield a serious crash. The next step is to perform the test on different potentially dangerous maneuvers.

Source: Phys.org

Chinese Go grandmaster Ke Jie, left, was defeated by the artificial intelligence program AlphaGo.

Google's AlphaGo has defeated world's number one Go player

As reported here at Neowin, AlphaGo defeated the current world number one Go player, Ke Jie, at DeepMind's "Future of Go Summit" in Wuzhen, China. The contenders have played three matches throughout the week, starting last Tuesday, and the AI-based program has managed to win all of them, marking "the highest possible pinnacle for AlphaGo as a competitive program".

Following this week's achievements, DeepMind has decided to retire AlphaGo from playing against humans and will publish a special set of 50 AlphaGo vs AlphaGo games "played at full length time controls" to inspire the Go community. The first ten games can be accessed here.